Turbo always boost all cores

Chike

Active Member
I don't know if it is normal or a bug in BIOS or HWiNFO, but all cores on my 4670k boost to max even if I run one thread with affinity to CPU 0 (or any other CPU)
 
According to Intel, 4th Generation Core family processors do not support each core running at different ratios.
 
There are several explanations for what you are seeing, or more than one of them combined that produces the result.

What you may be seeing with all cores running at the same Turbo frequency is a feature created by mother board manufactures called Multi Core Enhancement, or something similar to that. It keeps all the cores at the maximum Turbo multiplier. Some boards like Gigabyte have this feature enabled by default.

It could also be the BIOS setting for the CPU core ratios/multipliers, if it is anything besides an Auto or stock setting. For example, if the Core Ratio option is set to All Cores, all cores will be set to the same ratio/multiplier.

The standard way Turbo 2.0 works is the maximum Turbo frequency, determined by the core multiplier, depends on how many cores are being Turbo boosted. For example, the Xeon processor I've been using recently has these default multipliers:

1 Active Core 40
2 Active Cores 40
3 Active Cores 39
4 Active Cores 38

So normally during a CPU stress test, the maximum Turbo speed of each core is 3.8GHz, from the multiplier of 38. This is how Turbo works for Intel 2nd through 5th generation processors.

Intel's definition of an "Active" core is important, and is not what we would expect, given what I have seen using Intel's Extreme Tuning Utility.


I also see all my cores having a Maximum Core Clock in HWiNFO of 4000MHz, but that happens because each core at one time or another was Turbo boosted to 4000MHz, while the others were not being Turbo boosted. Since my processor can have two cores running with a multiplier of 40, it is more likely for me to have all of them recorded in the Maximum Core Clock field as 4000MHz.

In my UEFI/BIOS, I can choose the Per Core option for the CPU Ratio, and set the ratio/multiplier for each number of active cores. But that does not mean each core can be set to a different speed, and they can't.
 
In fact, by default my board does run 4 cores at max Turbo (x38) even under full load.
But this behavior is observed even if I set cores  manually to
1 Active Core 38
2 Active Cores 38
3 Active Cores 37
4 Active Cores 36
As per CPU specs

We do not have control on how many cores windows will use, but if only one load thread is running with affinity to core 0 or 1, I think it is safe to assume not more than 2 will be active, and still, all cores max frequency is 3800, but as Martin says, this is normal for 4th gen CPUs.
However it's just the clock that is high, if the load is on core > 0, vcore, that I assume is measured on core 0, only is at or close to it's minimum, which mean the core is not active, or there is very little load on it.
I thought it's normal but then someone posted a capture where you can see different cores with different max frequency.
Windows itself may have something to do with it because for the first few minutes all cores are at max even without a load and Balanced power profile is selected
 
Well, the screen posted might show different clocks across cores because the resolved core ratio/frequency can change several times per second. So while HWiNFO reads the current status from each core (starting from the first in a cycle) the other one might change the ratio during the process, which is then captured by HWiNFO. It's very difficult to report actual core clock because of this.
Also Intel's document are sometimes not clear about this. Somewhere I have seen claiming that 4th gen CPUs allow different ratios per core, but mostly they claim this is not supported.
The actual clock is also capped by several other factors, especially the power limiting logic (PL1, PL2, tau, cTDP, etc).
 
Martin said:
Well, the screen posted might show different clocks across cores because the resolved core ratio/frequency can change several times per second. So while HWiNFO reads the current status from each core (starting from the first in a cycle) the other one might change the ratio during the process, which is then captured by HWiNFO. It's very difficult to report actual core clock because of this.
Yes I know, I can sometime see one core is different from the others for the duration of one sample, but for the life of me I am not able to make a capture where two are different from the other two, let alone in this order.

Martin said:
Also Intel's document are sometimes not clear about this. Somewhere I have seen claiming that 4th gen CPUs allow different ratios per core, but mostly they claim this is not supported.
The guy has 3 different CPU which one of them is 3rd gen or less, need his feedback on which one was used for this capture.

Martin said:
The actual clock is also capped by several other factors, especially the power limiting logic (PL1, PL2, tau, cTDP, etc).
But those must work for the CPU and not per core don't they?
 
Examples of different core speeds I get in attachments. Windows 8 Snipping Tool makes this easy.

Why not have different ratios for each core, as long as they are at various levels between the maximum and minimum? I see mine do this all the time. Compare the VIDs shown for each core compared to the core speed, each core speed and VID will be different, when using Adaptive voltage and all CPU power saving options enabled.

It may help to adjust the Scan Interval in HWiNFO from the default 2000ms/2 seconds, to catch the cores at different speeds. The actual change in core ratios can happen so many times per second that a scan interval that makes sense for humans is super slow for a processor.

Chike, I also see all my cores running at the same maximum speed for about a minute or two after Windows boots, and then they go down after that. I see this on Ivy and Sandy Bridge systems too.

Since when does the UEFI/BIOS on any mother board that allows over clocking follow the rules of Turbo processor speed defined by Intel? ;)  The rules are broken even with my Haswell Xeon processor that cannot be over clocked with the core multipliers.
 

Attachments

  • hwinfo diff core 3.PNG
    hwinfo diff core 3.PNG
    26.4 KB · Views: 7
  • hwinfo diff core 2.PNG
    hwinfo diff core 2.PNG
    27.4 KB · Views: 7
It really does not bother me at all, the two Prime95 threads and one folding @ home don't leave my CPU much time to idle, I just thought it might be software bug, if it's not than it's cool.
Martin is doing the hard work, the least we can do is to report possible bugs.

BTW Snipping tool is not new in windows 8, and you could always capture the active window pressing alt + printscreen keys ever since I can remember (windows 3.1?).
 
Back
Top